Poker-AI.org

I have HUNL strategies for < 25BB based on pure CFRM. However, they don't perform well against human competition. I'm not a poker player so I can only loosely speculate that it's because they don't adapt to all-in ranges.

If anybody is interested in improving these strategies, I can share them with you. Just PM me here.

How were you able to conclude that? Is your sample number decent? Don't forget variance.

Also, CFRM provides a static defensive strategy. It isn't a best response strategy, so it is in no way the best strategy to use against humans that play an exploitable strategy, plus humans cannot be as consistent, increasing variance.



I think I would be interested in that as I am trying to achieve the same thing for my thesis. It would be a great help if we could discuss this.

Some background: To describe hand value, my HUNL strategy uses Hand Strength, (positive) Hand Potential, and a custom metric that measures river hand strength histogram homogeneity. I tried a range of other metrics when developing the abstractions and tested them against one another but these seemed to perform the best. Each stack level strategy was converged in increments for a few hours at a time then tested in intervals against a base strategy to avoid over-fitting. Game state and betting abstraction I've discussed elsewhere on this forum. Not much has changed in that regard.

I'm particularly interested in learning new techniques (of which I can later improve upon) or a new way of approaching the game.

Dear Nasher,

I'm one your neighbors that has been relentlessly stalking and harassing you for years now. I just wanted to say that on behalf of myself and the other mentally retarded psychotics in the area we are so incredible sorry and would like to award you $1.2m in damages for what has been gross misconduct on our part, including numerous felonies, and various attempts to fabricate evidence for the purpose of framing you.